Fiber-to-the-Room (FTTR) networks with multi-access point (AP) coordination face significant challenges in implementing Joint Transmission (JT), particularly the high overhead of Channel State Information (CSI) acquisition. While the centralized wireless access network (C-WAN) architecture inherently provides high-precision synchronization through fiber-based clock distribution and centralized scheduling, efficient JT still requires accurate CSI with low signaling cost. In this paper, we propose a deep learning-based hybrid model that synergistically integrates temporal prediction and spatial reconstruction to exploit spatiotemporal correlations in indoor channels. By leveraging the centralized data and computational capability of the C-WAN architecture, the model reduces sounding frequency and the number of antennas required per sounding instance. Experimental results on a real-world synchronized channel dataset show that the proposed method lowers over-the-air resource consumption while maintaining JT performance close to that achieved with ideal CSI, offering a practical low-overhead solution for high-performance FTTR systems.

The rapid development of the reconfigurable intelligent surface (RIS) technology has given rise to a new paradigm of wireless transmitters. At present, most research works on RIS-based transmitters focus on single-polarized RISs. In this paper, we propose a dual-polarized RIS-based transmitter, which realizes 4-transmit space-time block coding (STBC) transmission by properly partitioning RIS’s unit cells and utilizing the degree of freedom of polarization. The proposed scheme is evaluated through a prototype system that utilizes a fabricated dual-polarized phase-adjustable RIS. In particular, the polarization coupling phenomenon in each unit cell of the employed dual-polarized RIS is modeled and analyzed. The experimental results are in good agreement with the theoretical modeling and analysis results, and an initial research effort is made on characterizing the polarization coupling property in the dual-polarized RIS.

Traffic identification becomes more important, yet more challenging as related encryption techniques are rapidly developing nowadays. Unlike recent deep learning methods that apply image processing to solve such encrypted traffic problems, in this paper, we propose a method named Payload Encoding Representation from Transformer (PERT) to perform automatic traffic feature extraction using a state-of-the-art dynamic word embedding technique. By implementing traffic classification experiments on a public encrypted traffic data set and our captured Android HTTPS traffic, we prove the proposed method can achieve an obvious better effectiveness than other compared baselines. To the best of our knowledge, this is the first time the encrypted traffic classification with the dynamic word embedding has been addressed.